[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

Re: Recent s.s.t.c work



Original poster: Steve Ward <steve.ward@xxxxxxxxx>



Hi Ken,

 I think the noise-bursts on the upper
waveform signify the switching points; ideally they should occur at
the zero level.  (It would seem that the sync in the scope was not
fully tracking the upper waveform, so the switching points in the two
waveforms do not perfectly line up.)


Exactly. You can almost make out the bursts of noise in this primary current waveform captured from my medium sized DRSSTC:


<http://www.stevehv.4hv.org/DRSSTC1/scopepics/sparks2.JPG>http://www.stevehv.4hv.org/DRSSTC1/scopepics/sparks2.JPG

It takes a trained eye, but you can see where the trace thins out slightly and there is a minor distortion. That is set for 200A/div, you can see im hard switching about 180A on the last cycle... this is why low operating frequency is so desirable, it makes the driver delays less critical.


Clearly, then, I have to shift that phase before I fire it up with much more power.


If possible, that would be the best way to go. You could also possibly run it from the VCO, but that might be worse than just using the feedback with all the delays.



I had been starting to wonder why I did not just utilize feedback from the secondary rather than from the primary, as I did in my 1st t.c. But now I can see that one can't do that if one wants to keep the switching at zero-current. Zero current in the secondary is almost never zero current in the primary.


My first DRSSTCs used secondary feedback. It worked, but the reliability was not so great. After moving over to primary feedback i have not lost a single IGBT in DRSSTC use!!


Secondary feedback does offer some interesting modes of operation, though. In one phasing, the primary current tends to lead the inverter, while the other phasing makes the primary current lag. We want the current to LEAD the voltage if we have to choose between the two. In this manner, we are not turning off any current, because by the time the IGBTS switch off, the current is conducting through its internal diode. Of course, at turn ON, there may be a minor shoot-through condition as that diode shuts off.

So, leading primary current is better than lagging current, but true ZCS is optimal.

If anyone has some ideas as to how to maintain zero-current switching
in a simple feedback-system, I'd like to hear them!  In the interim,
I study the problem.


Im afriad i know of no simple way to completely cure this problem. What i suggest is tracking down where the majority of the delay is entering the loop. In fact, you dont even need to run any primary current for most of this testing. Just compare the input signal vs the output signal at various stages in your circuit. Perhaps compare the delay from feedback input, to gate driver input. Then compare from gate driver input to IGBT gate voltage. Perhaps there is something simple (like unwanted capacitance, or slow diodes) that is adding a lot of delay that could be avoided. I have basically given up on achieving real ZCS for my coils, but in your case, i would definately try to improve things if possible. Ideally you would get the phase lag down to about 20 degrees, which would mean you are hard switching about 1/3 of the peak current, max.


Steve W., to answer your other questions:  1.  The max. input voltage
will be ~twice the peak out of the variac, which is connected to go
to ~140 V rms.  That yields ~390 V across the H-bridge.  I do
incorporate a cycle-by-cycle over-current protection circuit that
seems to work OK, but I have not set it yet (via a pot).


You are using 1200V devices, use that to your advantage! I would rather suggest applying 280VAC with a V-doubler. To keep the peak primary current in check, just use less drive cycles (4-16 cycles as i mentioned, though i only needed 7 cycles to generate the 12 foot sparks). This would be much better than running half the input voltage at twice the current.



2. The feedback input signal is the output of my 2nd current transformer (following the 1 ohm resistor I've mentioned) clamped by 4 diodes in series--parallel. In other words, a ~2.8 V p-p square wave. Paralleling those diodes is a 50 ohm resistor and that whole network refers to signal ground.


Question: why the extra 1:1 transformer? It is possibly adding un-wanted delay, check it out and see.



The capacitor of a Schmitt-trigger-gate oscillator connects to the "top" of that network rather than to ground, so that the oscillator provides my "pilot oscillator" signal with no spark but the feedback-signal takes over as soon as the primary starts to draw current, turning the oscillator into merely an amplifier for the duration of the spark event. That seems to work seemlessly.


Could this capacitive coupling result in a phase shift of your feedback input? Perhaps im imagining the circuit wrong, but i would look into that. Maybe just use a resistor to raise the impedance of your oscillator output, that way the feedback can still "over ride" but there is no capacitance (which may be indirectly referenced to gnd, making the load on the feedback partially capacitive).



There necessarily exist phase-shifts within the overall feedback circuit: In the 1:100 primary-current transformer, in the 1:1 transformer following that, in the Schmitt-oscillator/amplifier circuit, in the capacitive coupling to the IGBT driver-transformer's primary, and within that transformer itself. All of them add up, of course. I had earlier thought that an automatic shift in operation of the feedback loop from dead-on resonance would take care of that, but apparently not. So...I need a means to keep that switching spot-on. Suggestions, anyone...?


Steve Conner has done some work with PLL's. I personally avoid them, but i often believe in the simple or elegant solution if it exists.



Ken Herrick

P.S. If I may, I will piggy-back the following invitation, addressed
to the "2 Steves" and to any others who, in any language, bear the
name of Steve (Steven, Stephen, Steppen, Stephanie, Steffen,
etc.): If your profession is that of scientist, in contrast to
"merely" engineer (like me) or technician, >and< if you accept the
theory of evolution, go to <http://ncseweb.org>ncseweb.org and add your names to the
roster of those subscribing to the...Steve-o-Meter! Currently, 644
such "Steves" are on the list, thereby firmly (and with humour)
endorsing the theory of evolution.



Afraid that i will only amount to an engineer... though im not really anything yet as im still yet a sophomore in college.


Steve

KCH

Tesla list wrote:
>Original poster: Steve Ward
><mailto: steve.ward@xxxxxxxxx><<mailto:steve.ward@xxxxxxxxx>steve.ward@xxxxxxxxx>
>
>Hi Ken,
>
>On 10/14/05, Tesla list <<mailto: tesla@xxxxxxxxxx><mailto:tesla@xxxxxxxxxx>tesla@xxxxxxxxxx> wrote:
>Original poster: "K. C. Herrick" <<mailto:kchdlh@xxxxxxx><mailto:kchdlh@xxxxxxx> kchdlh@xxxxxxx>
>
>2 Steves (& others)-
>
>I >am< measuring (differentially) across just the 1-ohm.
>
>
>Steve Conner mentions that maybe a differential measurement is not
>the best? Maybe double check with a standard probe connection (one
>end of CT grounded, probe on other end). Im not too familiar with
>doing differential mode on my scope, so i just do it the normal way.
>
> Since the
>1:1 xfmr is in series, I discount it as affecting the current--which
>is a pretty good sine wave. See
><<http://www.hot-streamer.com/temp/KCH_TCH4.jpg>http://www.hot-streamer.com/temp/KCH_TCH4.jpg >http://www.hot-streamer.com/temp/KCH_TCH4.jpg
>
>
>
>Yes, that looks OK, but it seems you are getting some serious
>switching delays. I see 2 spots of noise on the sine wave
>half-cycle. Im guessing the first noise is the IGBTs shutting off,
>and then the second (larger) burst of noise is due to the other
>IGBTs turning ON? Dont suppose there is any way to check primary
>current vs maybe gate voltage or the bridge output V? Would require
>an isolated DC supply i suppose, but it wouldnt have to supply much
>power (100W at 100VDC should be adequate).
>
>top waveform (5 V/cm via 10:1 probes). The 4 m-ohm capacitor-common
>shunt wave is also shown (at 5 V/cm, directly connected); I connect
>to the shunt via a ~15 foot tw-pair which I've now (arbitrarily)
>terminated with 100 ohms at the scope. I set that up differentially
>also, and since that shunt is referenced to mains common, there's a
>bit of 60 Hz c.m. voltage there, which the scope seems not to mind.
>
>
>I dont understand what is going on in the bottom waveform? Is that
>the current supplied by the DC caps?
>
>The 1-ohm wave is about 4 V peak, implying 400 A peak primary current
>(at a relatively low mains-input from the variac). The other wave is
>across the (measured) 4 m-ohms; its zero-line is at the 2nd cm from
>the bottom so the voltage existing at the peaks of the upper wave
>appears alternately as ~1.8 and ~3.2. That implies peak currents of
>450 and 800 A, altho the negative peaks (when the primary drive
>reverses) are somewhat higher. So...that's a lot better than the
>wide disparity I seemed to see yesterday but still, somewhat at
>odds. So Steve Conner, you're no doubt right re current shunts. But
>this shunt is a commercial Janco item: 2 quite-short flat bars
>between a pair of substantial brass blocks. Not a whole lot of L
>there, it would seem, & the waveform seems not too different from
>what one might expect...right?
>
>Steve Ward, I measure 12 uH and 400 nF for the primary L & C,
>yielding a calculated Fr of 72 KHz, I believe.
>
>
> I did a quick sim in pspice using the 12uH and 400nF and 32
> cycles. With an extremely low primary resistance, you could get up
> near 7000A pk with only 400VDC input! With a realistic primary
> resistance of .1 ohms, the peak current hits 3.5kA at 32
> cycles! So your 1000A could be *very* real. The current soars
> this high because there is no secondary in place to pull that
> energy out. Also, this is why i suggest running some 5 to maybe 15
> cycles (now that i realize how low your tank impedance really
> is!). 32 cycles is almost definately too much, lets not speak of
> going higher! What is the planned input voltage? Even with 350VDC
> and running maybe running 10 cycles would probably put you at 1000A
> pk and probably 6-7 foot streamers. Running up to 700VDC, you
> could probably back down to 5-7 cycles and achieve 10' streamers
> with maybe 1000-1300A pk.
>
>So if your driver has survived at what we speculate is 1000A pk, and
>running 32 cycles (a rather long duration in my opinion) then if you
>had the secondary installed for that kind of use, id guess you would
>be seeing at least 6 foot streamers (probably more). First thing i
>would do is re-program your counter to operate from maybe 4-16
>cycles. When you start pushing high power levels, the difference of
>just 1 cycle might mean a 15% increase in spark length, so you want
>a fine control over the cycles.
>
>
> Not too far from the
>(eyeballed) 77. And I do use feedback from the primary's current
>xfmr to essentially set up an oscillator, during the spark event,
>with the primary as its resonant circuit. That seems to be working
>well.
>
>
>What does the feedback input look like (waveforms)? Are you getting
>a nice square wave back? Or are you amplifying a low level sine
>wave? I lost track of your feedback scheme. I mainly ask because i
>have a feeling that somehow its adding in more delay than necessary.
>
> I'm presently operating with a 32-cycle gate-duration; when I
>actually start to make some sparks, I'll consider changing the
>current selection range from 32/64/128/256/512 cycles to a lower one,
>as you have suggested.
>
>
>Well, i wouldnt put more than 100VDC into the system as it stands
>now, running that many cycles and especially without the loading you
>could be running serious currents.
>
>
>As to operating w/out the secondary...I seem to keep putting you
>(Steve W.) in the position of reminding me of things I learned 50 yrs
>ago & should still attend to--e.g. what limits the primary
>current. Only problem is, I have to put the whole apparatus on the
>floor when the secondary is on it, & my old bones really creak when &
>if I have to crouch down over it to check on this or
>that. Unhappily, my shop has only an 8 1/2 ft ceiling rather than
>the 20 ft one I really need these days. And outside, being somewhat
>on a hill, I have no handy flat area anywhere nearby.
>
>
>Well, best of luck. I'd say if you can somehow reduce your
>switching delay, and if the 1000A you measured is real, then this
>thing is definately ready for serious spark production!
>
>Steve
>
>KCH
>
>
>
>
>
>